Contents

Site by site output for linkage model.

When the SITEBYSITE option is chosen, the output file contains posterior population of origin assignments for each allele copy at each locus for each individual. For large datasets, this information can require several megabases to store. Each line shows the assignment probabilities for one locus for one individual.The first two columns of the line indicate the number of the individual (ranging from 1 to NUMINDS) and the number of the locus (ranging from 1 to NUMLOCI) in the order that they occur in the data file. If the data file contains labels for each individual or marker names for each locus, these are given in subsequent columns. The format of the posterior assignment probabilities depends on the parameter combinations. If LINKAGE=0 or PHASED=1 then the first K rows of output give the probability that the first allele copy at the locus comes from populations 1..K. For diploid or polyploid data, analogous probabilities for subsequent allele copies are shown in further columns. If the linkage model is used (LINKAGE=1) and the data is not fully phased (PHASED=0) the posterior assignment probabilites for the allele copies at each locus can be strongly co-dependent. Structure therefore outputs joint assignment probabilities for the two allele copies implying $K^2$ entries for each locus (note that this option is not available for PLOIDY$\neq2$). If MARKOVPHASE=1 then the first K columns give the probabilities that the first allele copy in the datafile is in population 1 and the second allele copy is in population 1..K, with subsequent columns relating to probabilities with the first allele copy in populations 2..K. If MARKOVPHASE=0, then instead of referring to the first and second listed allele copies in the data file, the probabilities refer to the population of origin of maternal and paternal strands. If there is no phase information (PHASEINFO=0), then the posterior probability matrix should theoretically be symmetric, such that the probability the maternal allele is in population $k_1$ and the paternal allele is in $k_2$ will be equal to the probability that the maternal allele is in population $k_2$ and the paternal allele is in population $k_1$. In practice, because MCMC is used to estimate the matrix, there will be noticeable deviations from symmetry if NUMREPS is small. For example, suppose that the below is site-by-site output for two loci for a diploid individual with no phase information, with MARKOVPHASE=0. 

1

1

locX1

Ind1

5.65E-4

2.18E-6

7.95E-3

2.16E-6

1.22E-5

7.07E-4

7.95E-3

7.89E-4

9.82E-1

1

2

locX2

Ind1

5.20E-4

1.47E-6

7.93E-3

3.25E-6

1.33E-5

6.91E-4

8.01E-3

7.97E-4

9.82E-1

Then in order to calculate the assignment probabilities of the maternal and paternal allele copies at for the first locus the numbers are summed as follows: 

locus 1	pop1	pop2	pop3		origin of maternal(X)
					chromosome
pop1	5.65E-04	2.18E-06	7.95E-03		8.52E-03
pop2	2.16E-06	1.22E-05	7.07E-04		7.21E-04
pop3	7.95E-03	7.89E-04	9.82E-01		9.91E-01
origin of paternal	8.52E-03	8.03E-04	9.91E-01
chromosome (missing)

In this example, the data is from an X chromosome of a male, so in fact the second allele copy is missing.